User acknowledges that it has reviewed the User Agreement and the Privacy Policy governing this site, and that continued use constitutes acceptance of the terms and conditions stated therein.

Contemporary Trends in Model Risk Management

By Saqib Jamshed, OCC Senior Vice President, Model Risk Governance
February 27, 2020

Saqib Jamshed

"Model risk still is an extremely important component of a firm's operational risk profile. It deserves the utmost attention at the C-suite level, the Board of Directors, and regulators."

This is the logical next step in the evolution of model risk management practices as the frameworks underpinning them become more mature. What follows are some of the contemporary trends in model risk management gleaned from my interactions with peers at several financial services firms.

Maturation of Model Risk Management Frameworks
Model risk management frameworks dealing with development, implementation, use and validation of models have achieved a high degree of maturity in most financial services firms. The focus has now shifted to augmenting these frameworks with ongoing performance monitoring and implementation guidelines so that validated models are constantly monitored and recalibrated if necessary.

Validation of Artificial Intelligence and Machine Learning Models
Artificial Learning (AI) and Machine Learning models are a natural outgrowth of the massive amounts of data produced and consumed in the financial services industry. These models oftentimes use concepts and implementation mechanisms that do not lend themselves to be easily validated by more traditional methods in an independent and effective manner. This has resulted in exploration of new methods of validation such as extensive deployment of challenger models or deployment of AI models in parallel with existing models with a view to improve their accuracy.

Use of stress period data for development and calibration of models
Some of the more common time periods used for back testing and calibration of models for stress scenarios such as 2007-08 or 1997-98 are becoming dated and use of hypothetical stress scenarios is increasing. Model validation departments need to be extra vigilant when results from such hypothetical scenarios are evaluated or replicated due to potential biases coming into play when these models are developed.

Representation of model risk in Risk Appetite reporting
Model risks have traditionally been hard to incorporate in Risk Appetite frameworks that are more geared towards market credit and other more standard operational risks. With the maturation of model risk frameworks, organizations are taking a more nuanced look at the associated risk appetite reporting process. While the vast majority if not the entire model inventory has already been validated in most financial firms, risk appetite frameworks require some well-thought out re-engineering efforts so that model risk is conveyed appropriately to senior managers, Boards of Directors, and regulators.

A more integrated approach towards model development and validation
Organizations have been striving for better alignment between model development and validation functions in order to reduce time taken by both activities and to foster a more integrated and collaborative model risk management culture. AI and machine learning models, as well as a more pronounced emphasis on ongoing monitoring and calibration of models also are driving the need for closer collaboration between the first and second lines. The challenge is to continue making progress in that endeavor without sacrificing the independence of the model validation function that results in an effective challenge of model development activities.

Onshore and offshore model development and validation
Organizations largely housed both model developers and validators onshore as these were extremely specialized and required a lot of interaction with regulators on a regular basis. As model risk management frameworks have matured and initial validation of existing model inventory is complete, there has been some shift of both model development and validation activities offshore. Companies are not necessarily moving to the lowest-cost locations, but to locations that have an abundance of quantitative talent and language competencies that can replicate some of other control function offshoring endeavors. These arrangements will continue to evolve as model development and validation-related activities become more standardized as a result of business process re-engineering or robotic process automation initiatives underway at most financial firms.

LIBOR replacement for financial contracts expiring beyond 2021
The London Interbank Offer Rate (LIBOR) is an interest rate used extensively in financial instruments such as swaps. Regulators have communicated the need to stop using LIBOR as a benchmark in financial contracts maturing post-2021 as the LIBOR submission mechanism would not be necessary for participating banks at that time. Any model that uses LIBOR in any capacity will likely need to be modified or replaced. Financial firms had been focused on remediation of more pressing legal issues associated with existing contracts that use LIBOR but are now pivoting to assessing remediation efforts needed in the modeling and validation domains.

There are several other noteworthy trends, such as changes in Dodd-Frank Act stress testing and consolidation in regulatory regimes across the globe, that deserve a more detailed mention. However, the underlying message is still the same: model risk still is an extremely important component of a firm's operational risk profile. It deserves the utmost attention at the C-suite level, the Board of Directors, and regulators.

Learn more about OCC's transformation on our website.

To learn more about OCC's thought leadership on our blog.


Category: Compliance, Operational Risk Management, Regulation, Risk Management